Maximum likelihood estimation is computatonally infeasible for latent
variable models involving multivariate categorical responses, in parti
cular for the LISCOMP model. A three-stage generalized least squares a
pproach introduced by Muthen (1983, 1984) can experience problems of i
nstability, bias, non-convergence, and non-positive definiteness of we
ight matrices in situations of low prevalence, small sample size and l
arge numbers of observed indicator variables. We propose a quadratic e
stimating equations approach that only requires specification of the f
irst two moments. By performing simultaneous estimation of parameters,
this method does not encounter the problems mentioned above and exper
iences gains in efficiency. Methods are compared through a numerical s
tudy and an application to a study of life-events and neurotic illness
.